朴素贝叶斯

算法优缺点

• 优点：在数据较少的情况下依然有效，可以处理多类别问题

• 缺点：对输入数据的准备方式敏感

• 适用数据类型：标称型数据

算法思想：

朴素贝叶斯
比如我们想判断一个邮件是不是垃圾邮件，那么我们知道的是这个邮件中的词的分布，那么我们还要知道：垃圾邮件中某些词的出现是多少，就可以利用贝叶斯定理得到。
朴素贝叶斯分类器中的一个假设是：每个特征同等重要

函数

loadDataSet()

创建数据集，这里的数据集是已经拆分好的单词组成的句子，表示的是某论坛的用户评论，标签1表示这个是骂人的

createVocabList(dataSet)

找出这些句子中总共有多少单词，以确定我们词向量的大小

setOfWords2Vec(vocabList, inputSet)

将句子根据其中的单词转成向量，这里用的是伯努利模型，即只考虑这个单词是否存在

bagOfWords2VecMN(vocabList, inputSet)

这个是将句子转成向量的另一种模型，多项式模型，考虑某个词的出现次数

trainNB0(trainMatrix,trainCatergory)

计算P(i)和P(w[i]|C[1])和P(w[i]|C[0])，这里有两个技巧，一个是开始的分子分母没有全部初始化为0是为了防止其中一个的概率为0导致整体为0，另一个是后面乘用对数防止因为精度问题结果为0

classifyNB(vec2Classify, p0Vec, p1Vec, pClass1)

根据贝叶斯公式计算这个向量属于两个集合中哪个的概率高

1.  1 #coding=utf-8
    2 from numpy import *
    3 def loadDataSet():
    4     postingList=[[‘my‘, ‘dog‘, ‘has‘, ‘flea‘, ‘problems‘, ‘help‘, ‘please‘],
    5                  [‘maybe‘, ‘not‘, ‘take‘, ‘him‘, ‘to‘, ‘dog‘, ‘park‘, ‘stupid‘],
    6                  [‘my‘, ‘dalmation‘, ‘is‘, ‘so‘, ‘cute‘, ‘I‘, ‘love‘, ‘him‘],
    7                  [‘stop‘, ‘posting‘, ‘stupid‘, ‘worthless‘, ‘garbage‘],
    8                  [‘mr‘, ‘licks‘, ‘ate‘, ‘my‘, ‘steak‘, ‘how‘, ‘to‘, ‘stop‘, ‘him‘],
    9                  [‘quit‘, ‘buying‘, ‘worthless‘, ‘dog‘, ‘food‘, ‘stupid‘]]
   10     classVec = [0,1,0,1,0,1]    #1 is abusive, 0 not
   11     return postingList,classVec
   12 
   13 #创建一个带有所有单词的列表
   14 def createVocabList(dataSet):
   15     vocabSet = set([])
   16     for document in dataSet:
   17         vocabSet = vocabSet | set(document)
   18     return list(vocabSet)
   19     
   20 def setOfWords2Vec(vocabList, inputSet):
   21     retVocabList = [0] * len(vocabList)
   22     for word in inputSet:
   23         if word in vocabList:
   24             retVocabList[vocabList.index(word)] = 1
   25         else:
   26             print ‘word ‘,word ,‘not in dict‘
   27     return retVocabList
   28 
   29 #另一种模型    
   30 def bagOfWords2VecMN(vocabList, inputSet):
   31     returnVec = [0]*len(vocabList)
   32     for word in inputSet:
   33         if word in vocabList:
   34             returnVec[vocabList.index(word)] += 1
   35     return returnVec
   36 
   37 def trainNB0(trainMatrix,trainCatergory):
   38     numTrainDoc = len(trainMatrix)
   39     numWords = len(trainMatrix[0])
   40     pAbusive = sum(trainCatergory)/float(numTrainDoc)
   41     #防止多个概率的成绩当中的一个为0
   42     p0Num = ones(numWords)
   43     p1Num = ones(numWords)
   44     p0Denom = 2.0
   45     p1Denom = 2.0
   46     for i in range(numTrainDoc):
   47         if trainCatergory[i] == 1:
   48             p1Num +=trainMatrix[i]
   49             p1Denom += sum(trainMatrix[i])
   50         else:
   51             p0Num +=trainMatrix[i]
   52             p0Denom += sum(trainMatrix[i])
   53     p1Vect = log(p1Num/p1Denom)#处于精度的考虑，否则很可能到限归零
   54     p0Vect = log(p0Num/p0Denom)
   55     return p0Vect,p1Vect,pAbusive
   56     
   57 def classifyNB(vec2Classify, p0Vec, p1Vec, pClass1):
   58     p1 = sum(vec2Classify * p1Vec) + log(pClass1)    #element-wise mult
   59     p0 = sum(vec2Classify * p0Vec) + log(1.0 - pClass1)
   60     if p1 > p0:
   61         return 1
   62     else: 
   63         return 0
   64         
   65 def testingNB():
   66     listOPosts,listClasses = loadDataSet()
   67     myVocabList = createVocabList(listOPosts)
   68     trainMat=[]
   69     for postinDoc in listOPosts:
   70         trainMat.append(setOfWords2Vec(myVocabList, postinDoc))
   71     p0V,p1V,pAb = trainNB0(array(trainMat),array(listClasses))
   72     testEntry = [‘love‘, ‘my‘, ‘dalmation‘]
   73     thisDoc = array(setOfWords2Vec(myVocabList, testEntry))
   74     print testEntry,‘classified as: ‘,classifyNB(thisDoc,p0V,p1V,pAb)
   75     testEntry = [‘stupid‘, ‘garbage‘]
   76     thisDoc = array(setOfWords2Vec(myVocabList, testEntry))
   77     print testEntry,‘classified as: ‘,classifyNB(thisDoc,p0V,p1V,pAb)
   78     
   79     
   80 def main():
   81     testingNB()
   82     
   83 if __name__ == ‘__main__‘:
   84     main()